>For cq, there can only be one outstanding event of each kind at a time. >So you can just keep a mask of events that happened, avoiding any mallocs. >Is that not the same for cm?
For a listening cm_id, there could be multiple connection requests associated with the listen in process at the same time. You also need to report additional information versus just indicating which event occurred. >Anyway, once you can get all events from a given object, you can just go >and remove all events, without state, reference counting or whatever. > >Thats what I advocate instead of waiting until all events are delivered. >Makes sense? I don't think that it avoids the issue. Are you referring to setting a bit in a kernel or userspace structure? I just don't see how this helps indicate to a userspace app that an event has occurred, provide them the necessary information regarding the event, plus give them some context so that they can associate the event with something. - Sean _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
